Professional Documents
Culture Documents
Keywords: Computational Fluid Dynamics (CFD) has gained a lot of popularity in the consequence assessments of gas
Gas explosion explosions in the past couple of decades. This work reviews Porosity/ Distributed Resistance (PDR) approach-
Flame acceleration based CFD solver (PDRFoam) for gas explosions. The PDR approach is used to model the effect of small-scale
CFD modelling
obstacles and only solve for the large scale congestion. PDR-based modified Favre averaged equations for mass,
momentum, enthalpy, and regress variable are solved using PDRFoam solver, which is developed as a new
application in OpenFOAM. The evaluation of the solver is done against medium-scale standard benchmark
experiments. Further, the effect of block-age ratio, different types of fuels, and different obstacle diameters
is explored. Using PDRFoam simulations, the pressure–time series, the flame arrival times, and the flame
speeds are predicted and the same are compared with experiments. Compared to standard body-fitted grid
simulations, the PDR approach is computationally economical; The predicted results are in good agreement
with experiments.
1. Introduction these studies can help to evaluate gas explosion solvers and models.
For example, experimental studies based on gas explosions are done
Accidents involving gas explosions cause significant damage to any to quantify the effect of flame propagation (Arntzen, 1998; Rogers,
gas storage or production plants. Examples of major accidents include 1995). The effect of confinement and congestion of the geometry
Flixborough (1 June 1974), Piper Alpha (6 July 1988), Buncefield (11 is investigated by modifying the bundles of obstacles in a confined
December 2005), Deepwater Horizon (20 April 2010), and Fukushima framework (Vianna and Cant, 2010; Bjerketvedt et al., 1997; Skjold
Daiichi (11 March 2011). The accidental release of gases has become et al., 2013). This is attributed to the disparate length scales of the
a repeated occurrence in the petroleum industry (Marsh, 2014). In the obstacles and establishment in the industry. Hisken et al. (2016) have
case of accidental release, a large volume of hydrocarbon gas (fuel) reported experimentally the impact of suppressing vortex shedding on
mixes with air (oxidizer) to form a flammable hydrocarbon-air mixture. the gas explosion overpressure generation. It was observed that the
If these flammable mixtures get ignited (due to any ignition source), vortex suppression can reduce the maximum overpressure by 20%–25%
an explosion might happen, leading to severe consequences in terms of in small-scale combustion.
injuries, fatalities, business interruptions, and asset damage. However, conducting experiments in large scales like a whole pro-
Quantitative Risk Analysis (QRA) studies are typically used to esti- cess plant requires a large area along with concerns about safety issues
mate the risks associated with the operation of an engineering process.
while dealing with hazardous gases. Moreover, the cost of conducting
It is used to support the understanding of the risk exposure to the em-
experiments for such scenarios is also huge. Therefore, many prac-
ployees, the environment, the company assets, and its reputation. The
tical problems consider empirical or CFD models for the estimation
risk assessment also helps to make cost-effective decisions and manages
of overpressure generation. The performance of empirical relations
the risks for the entire asset life cycle. Considering the consequences of
has not been consistent for all types of geometries and fuels (Lak-
such gas leaks, industries tend to do a lot of QRA studies. One of the
shmipathy et al., 2019). In the present investigation, the PDRFoam
most critical consequences of gas release is ignition of fuel–air mixture
tool (an open-source variant of OpenFOAM) is evaluated against a
leading to explosion. Tools are required to estimate the overpressure
database of experimental results. Different small-scale and medium-
generated during such explosions. Consequently, many experimental
studies are performed to understand the physics behind gas explosions; scale experimental investigations are used for the validation of the
∗ Corresponding author.
E-mail address: vagesh@iitm.ac.in (V.D. Narasimhamurthy).
1
Equal contribution.
https://doi.org/10.1016/j.jlp.2023.105164
Received 24 February 2023; Received in revised form 1 June 2023; Accepted 27 August 2023
Available online 30 August 2023
0950-4230/© 2023 Elsevier Ltd. All rights reserved.
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
2.1. Governing equations and PDR terms The standard 𝜅 − 𝜖 turbulence model is used with modified PDR
terms. Eqs. (9) and (10) are modified transport equations for turbulent
Eqs. (1), (2), (7), and (8) represent the PDR-based modified gov- kinetic energy (𝜅) and turbulent dissipation rate (𝜖) respectively.
erning (Favre averaged) equations for mass, momentum, enthalpy (un- [ ( ) ]
burnt and burnt) conservation and regress variable. The standard 𝜅 − 𝜖 𝜕 𝜕 𝜕 𝜇 𝜕̃
𝑘
̃
(𝛽𝑣 𝜌̄𝑘)+ (𝛽𝑗 𝜌̄̃
𝑘̃𝑢𝑗 ) = 𝛽𝑗 𝜇 + 𝑡 ̄𝜖 (9)
+(𝛽𝑣 𝑃 +𝑃𝑅 )−𝛽𝑣 𝜌̃
model (Launder and Sharma, 1974) is used along with PDR-based 𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜎𝑘 𝜕𝑥𝑗
modifications for turbulence modelling. Using the classical gradient [ ( ) ]
hypothesis, turbulent terms in Eqs. (7) and (8) are modelled. 𝜕 𝜕 𝜕 𝜇 𝜕̃
𝜖 𝜖̃ 𝜖̃2
(𝛽𝑣 𝜌̃
̄𝜖 ) + (𝛽𝑗 𝜌̃ 𝑢𝑗 ) =
̄𝜖 ̃ 𝛽𝑗 𝜇 + 𝑡 + 𝐶1 𝛽𝑣 𝑃 − 𝛽𝑣 𝐶2 𝜌̄
𝜕 𝜕 𝑚̄ 𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜎𝜖 𝜕𝑥𝑗 ̃
𝑘 ̃
𝑘
(𝛽 𝜌)
̄ + (𝛽 𝜌̃
̄𝑢 ) = (1)
𝜕𝑡 𝑣 𝜕𝑥𝑖 𝑗 𝑖 𝑉 𝜖̃
+ 𝐶1 𝑃𝑅 (10)
̃
𝑘
𝜕 𝜕 𝜕 𝑝̄ 𝜕
(𝛽 𝜌̄𝑢̃ ) + (𝛽 𝜌̄𝑢̃ 𝑢̃ ) = −𝛽𝑣 𝛿 + (𝛽 𝜏̃ ) + 𝛽𝑣 𝜌𝑔
̄ 𝑖 Here, 𝜇 is dynamic viscosity, 𝜇𝑡 is eddy viscosity, 𝜎𝑘 , 𝜎𝜖 , 𝐶1 and 𝐶2
𝜕𝑡 𝑣 𝑖 𝜕𝑥𝑗 𝑗 𝑖 𝑗 𝜕𝑥𝑖 𝑖𝑗 𝜕𝑥𝑖 𝑗 𝑖𝑗 are model constants. P is the turbulent kinetic energy production rate
𝜕
− 𝑅𝑖,𝑜 − (𝛽 𝜌̄𝑢̃′′ 𝑢′′ ) (2) term, and 𝑃𝑅 is the turbulence rate generated by sub-grid obstacles.
𝜕𝑥𝑗 𝑗 𝑗 𝑖 𝑃𝑅 is an extra source term from the PDR method which is empirically
Here, all the Favre-averaged terms are represented with an overhead modelled and depends on the length scale of obstacles (Puttock et al.,
tilde (for example ̃ 𝑢 is Favre averaged velocity). 𝜌 is the density, 𝑢𝑖 2022). All other constants in the model are taken with their standard
is the fluid velocity in 𝑖th direction, 𝜏𝑖𝑗 is stress tensor, 𝑝 is pressure, values (Launder and Sharma, 1974).
𝑔𝑖 is gravity vector and 𝑢′′ ′′
𝑖 𝑢𝑗 is the Reynolds stress tensor. The source
term for mass is represented as 𝑉𝑚̄ . The terms 𝛽𝑣 and 𝛽𝑗 represent the 2.3. Combustion modelling
volume porosity and area porosity of a cell which are PDR fields. The
term 𝑅𝑖,𝑜 is the drag produced by sub-grid obstacles, which is modelled Combustion is modelled using a laminar flamelet approach (Poinsot
by Eq. (3). and Veynante, 2005) as in most of the premixed cases. This is based on
1 the assumption that the probability of finding the reaction occurring
𝑅𝑖,𝑜 = ̄ 𝐴𝑢 |𝑢 |
𝜌𝐶 (3)
2 𝐷 𝑖 | 𝑖| is much less than the probability of finding burnt or unburnt gas.
Here, A is the wetted surface area of an obstacle and 𝐶𝐷 is the drag This implies that the flame thickness is very small as all the chemical
coefficient. Calculating 𝐶𝐷 for the approximation/modelling of drag reactions happen in the flame. Also, the smallest turbulent length scale
induced by sub-grid obstacles is an important step in the PDR approach. or eddy size is larger than the flame thickness. In such a scenario,
The current model assumes that the flow velocity (U) is uniform across combustion progress is modelled in terms of progress variable (c) or
the grid. Local Reynold’s number (Re) is calculated using the length regress variable (b). Eq. (11) represents the mathematical definition of
the progress variable. The Favre averaged regress variable (𝑏)̃ is used
scale of obstacle (D), i.e. 𝑅𝑒 = 𝑈𝜈𝐷 . Using the local 𝑅𝑒 and the relations
given by Puttock et al. (2000) are used for 𝐶𝐷 calculation. Eq. (4) is in combustion modelling, and Eq. (12) represents governing equation
used for cylinders. For the value of 𝐿𝑜𝑔10 (𝑅𝑒) in between 5.3 to 5.5, for the regress variable.
linear interpolation is used. 𝑇 − 𝑇𝑢 𝜌 − 𝜌𝑢
{ 𝑐= =1−𝑏= (11)
𝑇𝑏 − 𝑇𝑢 𝜌𝑏 − 𝜌𝑢
1.0 𝐿𝑜𝑔10 (𝑅𝑒) < 5.3
𝐶𝐷 = (4) Here 𝑇𝑢 and 𝑇𝑏 represent the temperature of unburnt & burnt gas. 𝜌𝑢
0.7 𝐿𝑜𝑔10 (𝑅𝑒) > 5.5
and 𝜌𝑏 represent the density of unburnt & burnt gas respectively.
The relations for sharp-edged obstacles have a similar form (Eq. (6)).
( ) ( )
Turbulent intensity (Eq. (5)) is also used in those relations. 𝜕 𝜕 𝜕 𝜕̃
𝑏 𝜕
𝛽 𝜌̄̃
𝑏+ ̄𝑢 ̃
𝜌̃ 𝑏= 𝐷𝑙 + 𝜌̄𝑢̃ ̃ ̃
′′ 𝑏′′ +𝛽 𝜌 𝑆 𝛯|𝛥𝑏|+𝛽𝑣 𝜌𝑢 𝑆𝐼
𝑈′ 𝜕𝑡 𝑣 𝜕𝑥𝑗 𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝑣 𝑢 𝑙
𝑈
𝐼= (5) (12)
𝐿𝑜𝑔10 (𝑅𝑒)
2
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
3
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
Fig. 3. Velocity magnitude contours and the corresponding streamlines : (a, e) flame in laminar regime, (b, f) flame interacts with the first row of obstacles, (c, g) flame interacts
with the second row of obstacles and is split into two halves, (d, h) flame interacts with the last row of obstacles.
Fig. 5. (a) Computational geometry (length - 9 m, width - 4.5 m, and height - 4.5 m)
and the obstacle arrangement for 20 pipes case. P1 is the ignition point. (b) pitch for
Fig. 4. Comparison of overpressure and flame surface area evolution with time. 20 pipes case, (c) pitch for 40 pipes case, (d) pitch for 80 pipes case, and (e) pitch for
56 pipes case.
4
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
Fig. 8. The comparison of the maximum overpressure obtained from the British gas
experiments with the PDRFoam predictions. The solid black line represents prediction
Fig. 6. Time trace of overpressure for 20, 40, 56, and 80 pipes. matching the experiments (while the two other dashed lines represent a factor of two
under- and over-prediction).
5
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
6
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
Fig. 10. Computational domain (length - 10 m, width - 6.25 m, and height - 8.75 m) and the obstacle geometry (diameter - 0.5 m and length - 8.75 m) for SOLVEX pipe bundle
case. P1 is the ignition point, which is the same for cases in (a), (b), and (c). (a) is two columns, (b) is the front column, and (c) is the rear column configuration.
Fig. 11. Time trace of overpressure in (a) two-column case, (b) front column case, and (c) rear column case. Flame arrival at different locations in vent direction in (d) two-column
case, (e) front column case, and (f) rear column case.
maximum overpressure significantly more compared to the lower con- 1997). PDRFoam can capture this trend qualitatively. For the same
gestion configuration. This also envisaged the fact that gases accumu- VBR, i.e. 0.2, we observe higher overpressure value for 4 × 4 configu-
late around the obstacles. So, more blockage means more accumulation ration (i.e. 16 pipes) compared to 2 × 2 configuration (i.e. four pipes)
of gases and hence, higher maximum overpressure values. Fig. 17(b) (Fig. 17(b)). Fig. 18 shows the comparison of maximum overpressure
presents the test cases for Sotra experiments at VBR = 0.2. The maxi- for all the test cases done for 3D Sotra experiments. A significant
mum overpressure for two different obstacle sizes (0.41 m, and 0.82 m) over-prediction can be seen for VBR = 0.5 with the experimental
is obtained in 17(b). The effect of obstacle size in the same VBR observations. This indicates that the PDR method is sensitive to VBR
(i.e. 0.2) is evaluated. In the literature (Bjerketvedt et al., 1997), for the as seen earlier also in the case of British gas simulations (Section 3.2).
same volume blockage ratio (VBR), the highest overpressure value is For higher congestion (or VBR), the PDRFoam generally shows over-
observed for the smallest diameter case. This can be because more shear prediction. Further investigation is required in the context of grid
layers are formed for smaller diameter geometry (Bjerketvedt et al., generation (and grid guidelines are required) for various geometries.
7
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
Fig. 12. Time trace of overpressure for (a) methane as a fuel and (b) propane as a fuel. In both (a) and (b), black represents two columns case, blue depicts the front column
case and the green depicts the rear column case. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 13. Flame surface depicted by iso-surface of regress variable, b = 0.5; at various
time instants for the rear column case with propane-air mixture.
Fig. 15. (a) Front and (b) plan view of geometry for 3D Sotra cubical vessel with
3 × 3 pipe bundle arrangement, P1 is the ignition point. The dashed lines are used to
show sides with polythene.
8
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
Fig. 16. The flame surface propagation at different instants of time corresponding to iso-surface of regress variable b = 0.5 for the 3 × 3, Volume Blockage Ratio (VBR) of 0.1
cases.
9
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
Fig. 19. The comparison of the geometric mean–variance (simulation/experiment) against geometric mean bias (simulation/experiment) for PDRFoam (b) The zoomed view for
the data points.
10
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
Fig. 23. The time trace of overpressure for various grids in SOLVEX rear column
propane case.
Fig. 21. Flame arrival at different locations in the vent direction (British gas case).
Fig. 24. The time trace of overpressure in different domains for the SOLVEX rear
column propane case.
Fig. 22. Flame arrival at different locations in the vent direction (3D Sotra case).
Table 2 thereby highly sensitive to the domain size (boundary condition) and
Various grid sizes used in SOLVEX rear column propane case. the grid size. Note that, in the previous study of Puttock et al. (2022),
𝛥𝑋∕𝐷 Overpressure (mbar) PDRFoam did not show consistent behaviour in some simulations (for
0.8 1329.24 example see Figure 15, Figure 24, Figure 28 in Puttock et al. (2022)).
0.4 1417.54
0.2 1412.21
References
obstacles; thereafter the time of pressure rise in 0.8D grid case is more Arntzen, B.J., 1998. Modelling of turbulence and combustion for simulation of gas
realistic and the coarse grid gives better results for this specific case of explosions in complex geometries. (Ph.D. thesis). Norwegian University of Science
and Technology, NTNU.
SOLVEX rear column with propane as fuel.
Bjerketvedt, D., Bakke, J.R., Van Wingerden, K., 1997. Gas explosion handbook. J.
hazard. mater. 52 (1), 1–150.
Influence of domain size Dhiman, M., Meysiva, V., Sathiah, P., Narasimhamurthy, V.D., 2022. Poros-
ity/Distributed Resistance (PDR) modelling in the CFD solver PDRFoam. In: Recent
The venting geometry in the SOLVEX case is different compared to
Advances in Applied Mechanics: Lecture Notes in Mechanical Engineering. pp.
all other cases in this manuscript. To analyse the sensitivity towards 503–519.
venting boundary condition (domain size), we simulated the SOLVEX Hisken, H., Enstad, G., Narasimhamurthy, V.D., 2016. Suppression of vortex shedding
propane case with an extended domain. The domain size is taken as and its mitigation effect in gas explosions: an experimental study. J. Loss Prev.
Process Ind. 43, 242–254.
five times the actual cloud size. A grid size of 0.8D is used here.
Hjertager, B., Saeter, O., Solberg, T., 1996. Numerical modelling of gas explosions:
The secondary peaks seen in the exact domain case diminished in a review. In: International Symposium on Hazards, Prevention and Mitigation of
the extended domain case (see Fig. 24). The SOLVEX propane case is Industrial Explosion, Netharlands, Vol, 2. pp. 77–91.
11
M. Dhiman et al. Journal of Loss Prevention in the Process Industries 85 (2023) 105164
Lakshmipathy, S., Skjold, T., Hisken, H., Atanga, G., 2019. Consequence models for Puttock, J., Yardley, M., Cresswell, T., 2000. Prediction of vapour cloud explosions
vented hydrogen deflagrations: CFD vs. engineering models. Int. J. Hydrogen using the SCOPE model. Journal of Loss prevention in the Process Industries 13
Energy 44 (17), 8699–8710, Special issue on The 7th International Conference (3-5), 419–431.
on Hydrogen Safety (ICHS 2017), 11-13 September 2017, Hamburg, Germany. Rogers, M., 1995. Project EMERGE measurement of turbulent characteristics and drag
Launder, B.E., Sharma, B.I., 1974. Application of the energy-dissipation model of loads on pipework obstacles in steady unidirectional flows. Br. Gas Res. Technol.
turbulence to the calculation of flow near a spinning disc. Lett. heat mass transfer GRCR 610.
1 (2), 131–137. Sha, W., Launder, B., 1979. Model for Turbulent Momentum and Heat Transport in
Marsh, 2014. The 100 largest losses 1974–2013—Large property damage losses in the Large Rod Bundles.[LMFBR]. Technical report, Argonne National Lab., IL (USA).
hydrocarbon industry. 23rd edition, London: Marsh global energy risk engineering. Sha, W., Yang, C., Kao, T., Cho, S., 1982. Multidimensional numerical modelling of
Narasimhamurthy, V.D., Hisken, H., Atanga, G., Skjold, T., 2015. Porosity/Distributed heat exchangers. J. Heat Transfer 104 (3), 417–425.
Resistance modelling for industrial CFD applications. In: 8th National Conference Skjold, T., Pedersen, H., Bernard, L., Middha, P., Narasimhamurthy, V.D., Landvik, T.,
on Computational Mechanics (MekIT 2015), CIMNE, Trondheim, Norway. pp. Lea, T., Pesch, L., 2013. A matter of life and death: validating, qualifying and
321–331. documenting models for simulating flow-related accident scenarios in the process
Patankar, S.V., Spalding, D.B., 1974. A calculation procedure for the transient and industry. Chem. Eng. Trans. 31, 187–192.
steady-state behavior of shell-and-tube heat exchangers. Heat Exch.: Design Theory Snowdon, P., Puttock, J., Provost, E., Cresswell, T., Rowson, J., Johnson, R., Masters, A.,
Sourceb. 155–176. Bimson, S., 1999. Critical design of validation experiments for vapour cloud
Poinsot, T., Veynante, D., 2005. Theoretical and Numerical Combustion. RT Edwards, explosion assessment methods. In: Proceedings of International Conference and
Inc.. Workshop on Modeling the Consequence of Accidental Releases of Hazardous
Puttock, J., Cresswell, T., Marks, P., Samuels, B., Prothero, A., 1996. Explosion Materials, San Francisco. pp. 831–849.
assessment in confined vented geometries. SOLVEX large-scale explosion tests and Vianna, S.S., Cant, R.S., 2010. Modified porosity approach and laminar flamelet
SCOPE model development. Proj. rep. Health Saf. Exec. OTO 96 (4). modelling for advanced simulation of accidental explosions. J. Loss Prev. Process
Puttock, J., Walter, F., Chakraborty, D., Raghunath, S., Sathiah, P., 2022. Numerical Ind. 23 (1), 3–14.
simulations of gas explosion using Porosity Distributed Resistance approach part- 1: Weller, H.G., Tabor, G., Jasak, H., Fureby, C., 1998. A tensorial approach to compu-
Validation against small-scale experiments. J. Loss Prev. Process Ind. 75, 104659. tational continuum mechanics using object-oriented techniques. Comput. phys. 12
(6), 620–631.
12